Abstract
Automated anatomical landmark detection in fetal facial 3D ultrasound is critical for craniofacial assessment, but remains challenging due to acoustic noise, occlusions, and fetal pose variation. We propose an EMD-regularized 3D residual U-Net for heatmap-based detection of 19 facial landmarks. The distance-weighted regularization improves spatial localization by penalizing probability mass far from the target landmark. Evaluated on 695 scans from 130 fetuses, the model achieves a mean distance error of 1.77 1.48 mm, outperforming current state-of-the-art and approaching inter-observer error levels. We further introduce 3D expected local accuracy curves to characterize local confidence and false-candidate behavior around target landmarks. These results suggest that EMD-regularized heatmap regression is a promising approach for automated fetal facial landmark detection in 3D ultrasound.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/PIPPI_016.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/PIPPI_016_supp.pdf
Link to Open Review
BibTex
@InProceedings{TanYus_EMDRegularized_MICCAISAT2026,
author = { Tanriverdi, Yusuf Baran AND Alomar, Antonia AND Rubio, Ricardo AND Sukno, Federico AND Piella, Gemma},
title = { { EMD-Regularized Heatmap Regression for 3D Ultrasound Prenatal Facial Landmark Detection } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17257},
month = {pending},
page = {pending}
}